Information Value

Cluster

Cluster chosen: 3 with (648985 rows)

Preload

Information Value

Prepare frame

Dictionary:

• numeric: 927 features (Numeric)
• categorical: 227 features (Categorical)
• date: 161 features (Date)
• NA_features: 128 features (d_NA)
• Mean_features: 128 features (d_Mean)
• Min_features: 128 features (d_Min)
• MinMin_features: 1 features (d_MinMin)
• Max_features: 128 features (d_Max)
• MaxMax_features: 1 features (d_MaxMax)
• Time0_features: 128 features (d_Time0)
• FirstLast_features: 2 features (d_First, d_Last)
• Timing_features: 128 features (d_Timing)
• TimeToExitStation_features: 128 features (d_TtoE_S)
• TimeToExitProduction_features: 1 features (d_TtoE_P)

Formulas:

• Mutual Information = I(X, Y) = H(X, Y) - H(X | Y) - H(Y | X), understand un-sided
• Conditional Mutual Information = I(X, Y | Z) = H(Z | X) + H(X) + H(Z | Y) + H(Y) - H(Z | X, Y) - H(X, Y) - H(Z) = I(X, Y) + H(Z | X) - H(Z | Y) - H(Z | X, Y) - H(Z), with guarantee of I(X, Y | Z) > 0 and being one-sided

Leak reference: IV = 2.843399 (super large!!!) - 3 splits: [-Inf, -1.5), [-1.5, 164), [164, +Inf]

IV scale:

• over 0.5 = suspicious predictor but possible (literally dominates everything else)
• 0.3-0.5 = strong predictor
• 0.1-0.3 = average predictor
• 0.02-0.1 = weak predictor
• under 0.02 = so weak you mostly cannot do anything with unless you transform them

Cluster chosen: 3 with (1183747 rows)

• Positive count: 3499 (50.8650 %)
• Negative count: 645486 (54.8478 %)

Look at the splits

Table of 5000 saved rows (lowest absolute inflation value = 0.1001046 ):

• variable = Variable
• class = Split
• pct_bin = Percentage of samples included in the split
• good = Count of non-defect in the split
• bad = Count of defects in the split
• bad_rate = Relative count of defects in the split vs population
• inflation = Inflation rate of sample count on rates (thresholded)
• pct_good = Percentage of non-defect in the split
• pct_bad = Percentage of defect in the split
• odds = Odds ratio of the split
• woe = Weight of Evidence of the split
• miv = Information Value of the split

Characterstic Plots of Features (Unfacetted)

Characterstic Plots of Features (Facetted)

Characteristic Plots of Splits (Unfacetted)

Characteristic Plots of Splits (Facetted)